Search

OBJECTIVES: In biomedical research, spin is the overinterpretation of findings, and it is a growing concern. To date, the presence of spin has not been evaluated in prognostic model research in oncology, including studies developing and validating models for individualized risk prediction. STUDY DESIGN AND SETTING: We conducted a systematic review, searching MEDLINE and EMBASE for oncology-related studies that developed and validated a prognostic model using machine learning published between 1st January, 2019, and 5th September, 2019. We used existing spin frameworks and described areas of highly suggestive spin practices. RESULTS: We included 62 publications (including 152 developed models; 37 validated models). Reporting was inconsistent between methods and the results in 27% of studies due to additional analysis and selective reporting. Thirty-two studies (out of 36 applicable studies) reported comparisons between developed models in their discussion and predominantly used discrimination measures to support their claims (78%). Thirty-five studies (56%) used an overly strong or leading word in their title, abstract, results, discussion, or conclusion. CONCLUSION: The potential for spin needs to be considered when reading, interpreting, and using studies that developed and validated prognostic models in oncology. Researchers should carefully report their prognostic model research using words that reflect their actual results and strength of evidence. ispartof: JOURNAL OF CLINICAL EPIDEMIOLOGY vol:157 pages:120-133 ispartof: location:United States status: published

In this letter, environmental monitoring by mobile robots is considered where expensive or time-consuming sampling has to be carried out in order to obtain a metamodel of the phenomenon investigated. Due to limited resources, often not only a limited number of samples can be taken, but also the cost and time of the traveled distance between the sample points must be considered. We present an adaptive sampling method that greatly reduces the robot’s travel costs for all common sampling criteria with minimal impact on model accuracy. This is achieved by predicting future sample points based on virtual sampling over a horizon in each iteration of the algorithm and suggesting a next sample point after a cost optimization. The algorithm is simulatively evaluated for application to global exploration and reconstruction of unknown phenomena on a variety of randomly generated phenomena. It is shown that our method vastly outperforms standard adaptive sampling.

Forensic feature-comparison examiners compare-or "match"-evidence samples (e.g., fingerprints) to provide judgments about the source of the evidence. Research demonstrates that examiners in select disciplines possess expertise in this task by outperforming novices-yet the psychological mechanisms underpinning this expertise are unclear. This article investigates one implicated mechanism: statistical learning, the ability to learn how often things occur in the environment. This ability is likely important in forensic decision-making as samples sharing rarer statistical information are more likely to come from the same source than those sharing more common information. We investigated 46 fingerprint examiners' and 52 novices' statistical learning of fingerprint categories and application of this knowledge in a source-likelihood judgment task. Participants completed four measures of their statistical learning (frequency discrimination judgments, bounded and unbounded frequency estimates, and source-likelihood judgments) before and after familiarization to the "ground-truth" category frequencies. Compared to novices, fingerprint examiners had superior domain-specific statistical learning across all measures-both before and after familiarization. This suggests that fingerprint expertise facilitates domain-specific statistical learning-something that has important theoretical and applied implications for the development of training programs and statistical databases in forensic science. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Statistical learning (SL) is assumed to lead to long-term memory representations. However, the way that those representations influence future learning remains largely unknown. We studied how children’s existing distributional linguistic knowledge influences their subsequent SL on a serial recall task, in which 49 German-speaking seven- to nine-year-old children repeated a series of six-syllable sequences. These contained either (i) bisyllabic words based on frequently occurring German syllable transitions (naturalistic sequences), (ii) bisyllabic words created from unattested syllable transitions (non-naturalistic sequences), or (iii) random syllable combinations (unstructured foils). Children demonstrated learning from naturalistic sequences from the beginning of the experiment, indicating that their implicit memory traces derived from their input language informed learning from the very early stages onward. Exploratory analyses indicated that children with a higher language proficiency were more accurate in repeating the sequences and improved most throughout the study compared to children with lower proficiency.

Credit risk assessment is at the core of modern economies. Traditionally, it is measured by statistical methods and manual auditing. Recent advances in financial artificial intelligence stemmed from a new wave of machine learning (ML)-driven credit risk models that gained tremendous attention from both industry and academia. In this paper, we systematically review a series of major research contributions (76 papers) over the past eight years using statistical, machine learning and deep learning techniques to address the problems of credit risk. Specifically, we propose a novel classification methodology for ML-driven credit risk algorithms and their performance ranking using public datasets. We further discuss the challenges including data imbalance, dataset inconsistency, model transparency, and inadequate utilization of deep learning models. The results of our review show that: 1) most deep learning models outperform classic machine learning and statistical algorithms in credit risk estimation, and 2) ensemble methods provide higher accuracy compared with single models. Finally, we present summary tables in terms of datasets and proposed models.

Compositionality is a central component of the human faculty for generalization and flexibility. However, the computations involved are poorly understood, especially in terms of their cognitive costs. On one hand, compositionality requires searching combinatorially large hypothesis spaces, raising issues of tractability. On the other hand, compositional representations afford efficient and compact compression. To shed light on the cognitive resource required for compositionality, we used a within-subject time pressure manipulation to study how participants navigated a series of mazes, generated using recursive operations over spatial primitives. We find evidence that behavior is guided by the use of primitives and abstract operations over them, where the degree of compositional structure increases performance and speeds up decisions. And while time pressure led to more random errors, it did not impair the capacity for compositionality. Rather, participants increased their reliance on reusing and recombining previous computations, suggesting a remarkable robustness of human compositional reasoning.

People’s thoughts and feelings ebb and flow in predictable ways: surprise arises quickly, anticipation ramps up slowly, regret follows anger, love begets happiness, and so forth. Predicting these transitions between mental states can help people successfully navigate the social world. We hypothesize that the goal of predicting state dynamics shapes people’ mental state concepts. Across seven studies, when people observed more frequent transitions between a pair of novel mental states, they judged those states to be more conceptually similar to each other. In an eighth study, an artificial neural network trained to predict real human mental state dynamics spontaneously learned the same conceptual dimensions that people use to understand these states: the 3d Mind Model. Together these results suggest that mental state dynamics explain the origins of mental state concepts.

Contemporary theories of attentional control state that information can be prioritized based on selection history. Even though theories agree that selection history can impact representations of spatial location, which in turn helps guide attention, there remains disagreement on whether nonspatial features (e.g., color) are modulated in a similar way. While previous work has demonstrated color suppression using visual search tasks, it is possible that the location corresponding to the distractor was suppressed, consistent with a spatial mechanism of suppression. Here, we sought to rule out this possibility by testing whether similar suppression of a learned distractor color can occur for spatially overlapping visual stimuli. On a given trial, two spatially superimposed stimuli (line arrays) were tilted either left or right of vertical and presented in one of four distinct colors. Subjects performed a speeded report of the orientation of the “target” array with the most lines. Critically, the distractor array was regularly one color, and this high-probability color was never the color of the target array, which encouraged learned suppression. In two experiments, responses to the target array were fastest when the distractor array was in the high-probability color, suggesting participants suppressed the distractor color. Additionally, when regularities were removed, the high-probability distractor color continued to benefit speeded target identification for individual subjects (E1) but slowed target identification (E2) when presented in the target array. Together, these results indicate that learned suppression of feature-based regularities modulates target detection performance independent of spatial location and persists over time.

Developmental Dyslexia (DD) is a condition in which reading accuracy and/or fluency falls substantially below what is expected based on the individuals age, general level of cognitive ability, and educational opportunities. The procedural circuit deficit hypothesis (PDH) proposes that DD may be largely explained in terms of alterations of the cortico-basal ganglia procedural memory system (in particular of the striatum) whereas the (hippocampus-dependent) declarative memory system is intact, and may serve a compensatory role in the condition. The present study was designed to test this hypothesis. Using Magnetic Resonance Imaging, we examined the functional and structural brain correlates of sequence-specific procedural learning (SL) on the serial reaction time task, in 17 children with DD and 18 typically developing (TD) children. The study was performed over 2 days with a 24-h interval between sessions. In line with the PDH, the DD group showed less activation of the striatum during the processing of sequential statistical regularities. These alterations predicted the amount of SL at day 2, which in turn explained variance in children's reading fluency. Additionally, reduced hippocampal activation predicted larger SL gains between day 1 and day 2 in the TD group, but not in the DD group. At the structural level, caudate nucleus volume predicted the amount of acquired SL at day 2 in the TD group, but not in the DD group. The findings encourage further research into factors that promote learning in children with DD, including through compensatory mechanisms.

A key debate in language learning centers on how people successfully learn the extension of a novel word, despite inherent ambiguity in the input. Across two studies, we tested whether learners reduce ambiguity about a word’s extension by actively sampling the environment. Adult participants were first shown ambiguous learning situations in which novel words were presented with a set of exemplars that were drawn from a subordinate-level category (e.g., Dalmatians), a basic-level category (e.g., dogs), or a superordinate-level category (e.g., animals). Learners then had the opportunity to sample the label of additional exemplars. Participants systematically adapted their sampling choices as a function of training. Moreover, participants varied in their sampling strategies, pursuing both confirmatory strategies (selecting exemplars similar to the training set) and constraining strategies (selecting exemplars that constrain the word’s extension). Overall, these findings show that learners spontaneously pursue sampling strategies that support generalizing novel word meanings.

Predictive processes and numerous cognitive, motor, and social skills depend heavily on sequence and statistical learning. The visuomotor Serial Reaction Time Task (SRTT) can measure these fundamental cognitive processes. In this review, we sought to answer whether repetitive transcranial magnetic stimulation (rTMS) is suitable for modulating visuomotor sequence learning and which factors mediate its efficacy. We systematically analyzed the eligible records (n = 17) that attempted to modulate the performance of the SRTT with rTMS. The purpose of the analysis was to determine how the following factors affected SRTT performance: (1) stimulated brain areas, (2) rTMS protocols, (3) stimulated hemisphere, (4) timing of the stimulation, (5) SRTT sequence properties, and (6) other methodological features. The primary motor cortex (M1) and the dorsolateral prefrontal cortex (DLPFC) were found to be the most promising stimulation targets. Low-frequency protocols over the M1 usually weaken performance, but the results are less consistent for the DLPFC. These inconsistencies (at least in part) may be due to the current frequency-based heuristics of rTMS effects developed for explaining local stimulation effects in the motor cortex. Although this review cannot provide an ultimate answer about the optimal stimulation and task parameters, it comprehensively discusses the behavioral effects of six factors that are crucial for designing future studies to modulate sequence learning with rTMS. Moreover, we conclude that synergistically combining functional neuroimaging with rTMS could adequately link the rTMS-induced network effects with the behavioral findings that are crucial to develop a unified cognitive model of visuomotor sequence learning.

A key aspect of understanding how children learn the meanings of words involves understanding how they mine different sources of information (e.g., observational, linguistic) in the service of learning. According to one dominant view, there exists a class of words (i.e., “hard words”; Gleitman et al., 2005) for which learning their meaning requires access to information beyond the observational contexts in which those words occur. Building upon previous work on this topic that employed the Human Simulation Paradigm, a paradigm commonly used for investigating vocabulary learning, the current study revisits the role of observational contexts for the acquisition of one class of hard words: nouns that denote non-basic level object categories (or “hard nouns”; see Kako, 2005). These data reveal that although observational contexts may not be sufficient to yield learning of precise hard noun meanings, they allow learners to extract systematic partial knowledge, knowledge that may lay a critical foundation for meaning acquisition.

A recent proposal argued a structure like X1’s-X2 is recursive if X1 and X2 are substitutable as the head; the head requirement avoids wrong generalizations like ‘*dogs chase cats chase rats’ since NP is not the head of NP1-V-NP2 although the NPs can be substitutable. To test the role of headedness in the distributional learning of recursion, we exposed 50 adults to one of two artificial languages: both allowed A1-B-A2 strings, with A1 and A2 substitutable; but the head was A2 in one language and B in the other. Non-head elements were optional, so the two languages allowed different linear strings. In the rating test, at predicted, participants from both conditions learned the head; and while there was no difference for ABA strings between conditions, suggesting they all learned linear substitutability, participants exposed to head substitutability (A-head) were more willing to allow recursion (ABABA) with both attested and unattested words.

Rule learning, that is the ability to deduce and generalize rules beyond the irrelevant characteristics of the original stimuli and learning contexts (Marcus et al., 1999), is present from birth (Gervain et al., 2008, 2012). It allows infants to build domain-general cognitive networks (Anderson, 1996), develop a vocabulary (Saffran et al., 1996), and learn the grammatical structures of their mother tongues (Bettoni et al., 2020; Kidd & Arciuli, 2016). Previous work suggests large heterogeneity in infants’ and children’s rule learning (Rabagliati et al., 2019). In this study, we extend the data collected as part of the ManyBabies project on rule learning (Visser et al., 2022) to 1) assess the development of infants’ rule learning across a 6-month interval, 2) explore one potential source of the heterogeneity in infants’ rule learning, namely their early contingent interaction experience, and 3) investigate how individual differences in early rule learning are associated with their language development longitudinally.

Drift diffusion models are used to model evidence accumulation in two-choice forced-tasks. The traditional approach to fitting Ratcliff’s standard drift diffusion model (where the drift and diffusion are constant) usually involves explicit modeling of the first passage time distributions of the upper and lower boundaries or likelihood approximations. We present the very first technique, to the best of our knowledge, that foregoes use of explicit modeling of the first passage time distributions with a random forest regressor. A random forest regression model that takes the first five moments of the response time distribution, and the upper boundary termination proportion, is used to predict the drift and diffusion parameters from response time data. A training set of response time samples of size 2500 from 121 distinct drift-diffusion pairs is used to train the random forest regressor. On a testing set of 10,000 distinct drift-diffusion combinations with response time sample sizes of 40, we find that our model surpasses techniques that make use of some form of analytical modeling of the first passage time distributions of the boundaries for prediction of the diffusion rate, but not the drift rate. We conclude that the application of machine learning to drift-diffusion modeling of empirical data is a topic worth further investigation.

Computational cognitive modeling is an important tool for understanding the processes that support human and animal decision-making. Choice data in sequential decision-making tasks are inherently noisy, and separating noise from signal can improve the quality of computational modeling. Currently, most models assume that noise is constant, or static, typically by including a parameter (e.g., uniform ε) to estimate the noise level. However, this assumption is not guaranteed to hold -- for example, an agent can lapse into an inattentive phase for a series of trials in the middle of otherwise low-noise performance. Assuming that noise is static could bias parameter and model identification. Here, we propose a new method to dynamically infer noise in choice behavior, under a model assumption that agents can transition between two discrete latent states (for example, attentive and noisy). Using four empirical datasets with diverse behavioral and modeling features, we demonstrate that our method improves model fit and that it can be easily incorporated into existing fitting procedures, including maximum likelihood estimation and hierarchical Bayesian modeling.

Through statistical learning (SL) we can pick up the statistical regularities characterizing the sensory environment. The proposed link between SL and language is one of the most illustrative examples of the significance of SL (e.g., Erickson & Thiessen, 2015; Romberg & Saffran, 2010; Siegelman, 2020). But what evidence is there to support this claim? This meta-analysis aims to provide a quantitative overview of the studies that tested this link empirically by correlating individuals’ SL performance with measures of language ability. We aim to answer the following research questions: (1) What is the overall weighted mean effect size of published studies that have tied inter-individual variability in SL performance to inter-individual variability in performance on language tasks? (2) Does the size of observed correlations vary as a function of (A) sample characteristics (e.g., age), (B) the type of task SL administered and (C) the test of language ability? (3) Does the data from published studies suggest the presence of a publication bias in this field of research?

In recent years, a number of methodological approaches has been developed that allow for identifying patterns in first language acquisition. In most cases, such methods have been used to analyze monolingual acquisition. In a global perspective, however, monolingual acquisition is the exception, and multilingualism is the norm. This raises the question to what extent such pattern-detection methods can be used for multilingual data as well, especially in the case of code-mixing, i.e. the use of more than one language in a single utterance. In this paper, we explore the potential of a data-driven approach to child language data proposed by McCauley & Christiansen (2019a), viz. chunk-based learning, to account for multilingual acquisition data. In particular, we compare the patterns found in the children’s code-mixed utterances with those in their monolingual utterances to test the hypothesis that both types of utterances are ultimately constructed from the same “building blocks”.

What predicts individual differences in children’s acquisition of consonant production across languages? Considerations of children’s development of early speech production have traditionally emphasized inherent physiological constraints of the vocal apparatus that speakers generally have in common (i.e., articulatory complexity). In contrast, we propose a statistical learning account of phonological development, in which phonological regularities of the ambient language guide children’s learning of those regularities in production. Across four languages (English, Spanish, Japanese, and Korean), we utilized recent meta-analytic dataset of age of consonant acquisition spanning 28 studies. High-density measures of children's ambient language environment from over 8,000 transcripts of speech directed to over 1,000 children were used to assess how well the frequency of consonants in child-directed speech predict the age of consonant acquisition. Our results suggest that both frequency and articulatory complexity are related to age of acquisition, with similar results found for English, Spanish, Japanese, and Korean. Consonants heard frequently by children tended to be incorporated into their production repertoires earlier and consonants heard less frequently are incorporated into production repertoires later in development. We discuss future directions that incorporate a statistical learning pathway towards learning to produce the sound patterns of the ambient language.

Children learn word meanings from their patterns of usage in their everyday input. This trivial statement is made more interesting by the fact that most patterns of word usage, even around language-learning children, are not particularly good at revealing word meaning. So, do children simply ignore much of their input and learn words primarily from the few instances of usage where word meaning is transparent? Or are there pieces of information in the sea of opaque word usage that would allow children to learn words slowly over time? In an adaptation of Gleitman and colleagues’ classic Human Simulation Paradigm (Gillette et al., 1999), the current study explores the kinds of input that contribute to learning. Our data suggest that the answer may depend on how “learning” is assessed.

Distributional models have been successfully used to model a wide range of semantic behaviors. One notable problem with distributional models is that they lack a way to distinctly represent different kinds of semantic relations within a single semantic space. Here, we propose that neural network languages models can sensibly be interpreted as representing syntagmatic (co-occurrence) relations using their input-output mappings, and as representing paradigmatic (substitutability) relations using the similarity of their internal representations. We test this proposal on three neural network architectures (SRNs, LSTMs, and Word2Vec) using a carefully constructed artificial language corpus. The sentences in the corpus have systematic relationships such that words belonging to categories that predict what other categories are allowed to occur in the same sentence. Using this corpus, we show that the models display interesting but understandable differences in their ability to represent these two kinds of relationships.

When faced with uncertainty, human observers maximize performance by integrating sensory information with learned task-relevant regularities. Does this behavior similarly occur in social settings? In this paper, we explore how reward-seeking behavior in an intergroup context is affected by readily available but task-irrelevant social information (in the form of group membership) when task-relevant reward information can be learned over time. Across two experiments, we show that participants learned and utilized task-relevant regularities to inform their choices. We also show that human observers are not universally biased towards utilizing social information in all settings––participants learned to disregard social information when not relevant to the task at hand. However, learning about the utility of social information (Experiment 2) had a long-term influence on observers’ ability to subsequently learn and utilize available sources of information. Real-world intergroup contexts typically encompass situations and stimuli that have been previously experienced by the observer. Our findings highlight the powerful influence of learning in such contexts.

Response inhibition—our ability to suppress actions—is achieved through coordinated reactive and proactive processes that may recruit different neurophysiological mechanisms. Here we adapted a two-step continuous performance task in which the decision to respond depends on a combination of an initial context cue and a subsequent target probe. Using transcranial magnetic stimulation (TMS), we mapped changes in corticospinal excitability and inhibition, providing indices of reactive and proactive processes that influence the state of the motor system in the lead-up to initiating or suppressing a response. We found distinct changes in corticospinal excitability at critical timepoints when participants were preparing in advance to inhibit a response (during the cue) and while inhibiting a response (during the probe). Motor system activity during early timepoints correlated with behavioural indices of proactive capacity and predicted whether participants would later successfully inhibit their response. Implications for theories of inhibitory control will be discussed.

Recent work suggests that people produce random sequences by tapping into a general ability to derive probabilistic inferences through sampling an internal belief distribution. In contrast, previous work has proposed that participants monitor the production of inherently deterministic sequences, changing strategy when randomness appears low. Experiments exploring this process have highlighted the role of task requirements (e.g., production pace, cognitive load), finding that randomness decreased when demands increased. However, assuming random generation stems from sampling for inference, increasing production rates might affect sequential dependencies and the overall sample distribution independently. For example, increased production rates might result in more predictable sequences but leave the overall distribution of the sample intact. Here, we assessed the effects of production rates on sequential statistics and overall distributional properties in a preregistered experiment in which participants produced people's lifespans. We discuss our results in terms of computational models of sampling for inference.

Paradigmatic associations are second-order associations where the items share a common context rather than being directly associated. Despite the importance of the structure in knowledge representation, the underlying mechanisms to form paradigmatic associations are not well studied. In the current study, we examined whether explicit attentional control is critical for forming paradigmatic associations. We used an implicit learning task, which limits the use of explicit attentional control, to see whether the associations can be formed without attentional control. Results showed evidence for learning, which implies that explicit attentional control may not be necessary for forming paradigmatic associations. We also used the n-back task to examine whether the ability to maintain information is critical for forming paradigmatic associations. Results did not provide evidence for the relationship between the two. We discuss the results in terms of the core mechanisms that may enable the formation of higher-order associations.

Our memories are stored not in isolation, but in richly structured networks of knowledge. How are individual memories represented within these knowledge structures, and how do new experiences become integrated into existing structures? To address these questions, we exposed participants to novel objects from a hierarchical category structure. After participants learned these objects to criterion, we introduced new objects from the same underlying structure. Participants learned the new objects much more quickly than the old set, suggesting that recently learned knowledge rapidly scaffolded the learning of new, related information. However, fine-grained analyses of participants’ memory errors suggest these newly learned objects may not be fully integrated into prior learned category structure, but rather are held separate. These results suggest that existing knowledge can facilitate the learning of new information in the absence of assimilation of that information. Ongoing work investigates how these dynamics unfold over time and with sleep.

In science, we use common graphical representations to indicate changes in events over time, independent of domain. Are children also sensitive to abstract patterns in the ways events change over time? In a series of four experiments, we show that young children (range: 48-84 months) distinguish different function families (Exp 1). Children can also distinguish specific function types within function families (e.g., between linear and sigmoid monotonic functions; Exp 2); nonetheless, they will group different function types within a family together, rather than with functions in a different family (Exp 3). Finally, we show that children’s sensitivity to functions is abstract, allowing them to match observable causes to verbal descriptions of their effects (Exp 4). These results suggest that although some aspects of function understanding, like learning how to interpret graphs, requires formal education, the ability to identify abstract functional relationships is intuitive and early emerging.

Biases about social groups emerge from a young age. This study examines whether statistically representative counterevidence – a randomly drawn sample from the social group – can change children’s attitudes and beliefs about minimally defined social groups. We found that 5- to 6-year-olds learned from the sample to change their attitudes and beliefs about minimal groups. However, they showed a negativity bias and an ingroup bias when they learned from the evidence. It was unclear whether 9- to 10-year-olds’ attitudes and beliefs can also be changed by this type of evidence. Future directions and implications of this study are discussed.

Despite the importance of risk attitude on decision making, its role on belief updating has been overlooked. In this paper, we aim to answer this fundamental question. We show, using economic theory, that stronger risk aversion drives more conservative actions, decreases the instrumental value of information relative to the importance of belief-based utility. Thus, with utility for clarity, stronger risk attitude yields to more belief change; whereas with updating cost, stronger risk attitude leads to less belief change. We confirm the result experimentally in two settings where subjects update their belief about their IQ and a randomly drawn number, respectively.

Our group recently found that talker's own speech productions are affected by statistical information in another talker's speech acoustics. While it is believed that dimension-based statistical learning in speech perception drives these changes in production, the mechanism underlying this "transfer" of statistical learning is not yet known. It is possible that the mechanisms underlying this transfer are dependent on auditory feedback from one's own voice. If so, then it follows that a lack of overt speech production entails no transfer of statistical learning from speech perception to speech production. In this study, we test this hypothesis using a dimension-based statistical learning experiment. Participants passively listened to a short sequence of acoustically variable minimal pair (beer/pier) speech segments conveying either an accent or typical English acoustics, make a perceptual judgment to an ambiguous Test stimulus and then repeat the Test stimulus. In the Canonical condition, F0 and voice onset time (VOT) of /b/ and /p/ covary according to their usual pattern in English. In the Reverse condition, the correlation is reversed to create an ’accent’ with speech input regularities atypical of English. To test our hypothesis, the proportion of repeated Test stimuli differs according to which group the participants are assigned to. In the 100% Repetition Proportion group, participants repeat Test stimuli on every trial whereas in the 10% Repetition Proportion group, participants repeat Test stimuli only on every tenth trial. If only those participants assigned to the 100% repetition group show transfer of statistical learning from perception to production, it follows that auditory feedback of one's own voice is required for statistical learning to transfer to speech production.

Effectively updating one’s beliefs requires sufficient empirical evidence (i.e., data) and the computational capacity to process it. Yet both data and computational resources are limited for human minds. Here, we study the problem of belief updating under limited data and limited computation. Using information theory to characterize constraints on computation, we find that the solution to the resulting optimization problem links the data and computational limitations together: when computational resources are tight, agents may not be able to integrate new empirical evidence. The resource-rational belief updating rule we identify offers a novel interpretation of conservative Bayesian updating.

Humans can successfully interpret images even when significant image transformations have distorted them. Such images aid in differentiating existing computational models for perception because models which predict similar results for typical stimuli may diverge when confronted with atypical stimuli. We propose an explanation of people’s ability to perceive a specific class of degraded stimuli that require extreme generalization capabilities: Mooney, or two-tone images of faces. We provide a parametric and controllable method to generate Mooney faces and hypothesize that the human visual system generalizes under extreme cases because it reverses the naturalistic image generation process. We show that Mooney faces are approximated under extreme illumination conditions; a computer graphics perspective of the Mooney formation process suggests inverse graphics as a model for perceptual generalization. With a number of demonstrations, we argue that Mooney faces are important to understand the role of generative models and inverse graphics in human visual processing.

Uncovering how categories develop through childhood is crucial for cognitive science. However, even for simple domains, categories can be complex, making it challenging to access them experimentally, especially in developmental studies. Markov-Chain Monte Carlo with People (MCMCp) is a statistically-based procedure that allows us to elicit category members from participants' implicit categories. However, due to the complexity of the paradigm, MCMCp has been limited to experiments with adult populations. Here, we develop and validate a child-friendly method for applying MCMCp, producing the first MCMCp experiment to elicit category examples from children. Comparing fruit category members for five-year-olds and seven-year-olds, we find generally consistent representative fruits and a developmental progression of initially broad and overlapping fruit categories to more differentiated distributions.

This study will examine the efficacy of Vocabulary Acquisition and Usage for Late Talkers Treatment (VAULT) in a version where target words are selected according to their phonological properties. Specifically, researchers will analyze the phonological characteristics of words already in specific toddlers' lexicons, then the toddlers will receive VAULT in one of two conditions. In Condition 1, new words will have more familiar sounds; in Condition 2, new words will have less familiar sounds.

Orthographic regularity (OR), the statistical regularity of letter strings, is a critical factor in fluent visual word recognition. Although studies showed that this regularity could be learned by simple exposure to letter strings, it remains unclear whether OR could be learned when participants are unconscious of their exposure to letter strings. In this study, we compared OR learning between unmasked and masked (continuous flash suppression; CSF) exposure conditions. From Chetail (2017)’s study, regularities are embedded in the initial position of letter strings (e.g., A B _ _ _). Nine of fifteen participants who were slightly aware of the presentation of stimuli in CSF condition (< 30 %) significantly selected the letter stimuli embedded in OR as word-like both in CSF (69.2%) and non-CSF (69.9%) conditions, showing no significant difference between conditions. This result suggests that orthographic regularity can be learned even when letter strings are perceived unconsciously.

While there is evidence that bilingualism enhances statistical learning and substantially reduces the degree to which learners use mutual exclusivity (ME) constraints, little is known about the role of multilingualism. In this study, we tested ME in monolinguals, sequential-bilinguals, trilinguals, and quadrilinguals using a cross-situational statistical learning task. Participants were familiarized with a mixed set of one-word-to-one-object mappings and two-objects-to-one-word mappings in three consecutive phases, each of which was followed by a test. Results revealed that all language groups learned both mapping types by the end. They also learned more one-to-one mappings than two-to-one mappings. Inconsistent with previous research, bilinguals and monolinguals showed a similar learning trajectory of two-to-one mappings. However, both trilinguals and quadrilinguals outperformed monolinguals. Trilinguals did not differ from quadrilinguals in accepting multiple referents for the same label. Findings suggest that multilingual language experience might enhance cross-situational statistical word learning and ME relaxing ability more than bilingualism.

A central challenge of cross-situational word learning is retaining word-referent mappings across exposures. We evaluate Memory-Bound Pursuit (MBP), a hypothesis-testing model of cross-situational word-learning which aims to account for learners’ memory constraints via a single parameter targeting the number of words that can be learned concurrently. Here, we show that by varying this parameter with age, MBP can capture both children’s and adults’ cross-situational word-learning success under varying levels of ambiguity. We also present new experimental findings supporting novel predictions made by MBP about the retention of word-referent mappings across intervening exposures. These findings suggest that MBP provides a strong baseline model of cross-situational word learning, capturing both developmental trends and experimental evidence of memory limitations for word learning.

The ability to extract regularities from the environment, known as statistical learning (SL), facilitates the processing of complex stimuli. At the same time, metamemory studies have shown that people’s assessments of their own learning and memory – judgments of learning (JOLs) – are based on cues that pertain to the study materials (e.g., word frequency), study conditions (e.g., number of study opportunities) or that signal ease during the learning process (e.g., fluency). However, whether linguistic structures acquired via SL are used as cues to inform JOLs has not yet been addressed. In this study, we will test the impact of regularities in the co-occurrences of syllables within artificial words on JOLs. N = 90 participants will be exposed to a continuous auditory stream of artificial words with fixed transitional probabilities between adjacent syllables during a familiarization phase. Afterwards, they will study and make JOLs for items that follow the transitional probabilities and were presented in the familiarization phase (“word” item type), for items that follow the transitional probabilities but were not presented in the familiarization phase (“phantom” item type) and for items that do not follow the transitional probabilities and were not presented in the familiarization phase (“non-word” item type). To assess SL, one group of participants “SL-and-JOLs” (n = 45) will be asked whether the item belongs to the language from the auditory stream or not before making each JOL. To test SL influences on metamemory without prompting participants to make SL classifications, the other group of participants “JOLs-only” (n = 45) will be asked to make JOLs only. Finally, participants will complete a recognition memory task. We expect that if SL occurs, the proportion of items classified as words belonging to the language from the auditory stream will be higher for “word” and “phantom” items than for “non-word” items (SL-and-JOLs group only). Further, if SL occurs and the linguistic regularities acquired via SL inform JOLs, we expect that JOLs will be higher for “word” and “phantom” items than for “non-word” items in both groups. This study will shed light on whether cues for metamemory judgments can be learned through the extraction of regularities from the environment. This would indicate that SL does not only facilitate learning at an object-level, but that people also base their metamemory on its facilitating effects.

Recent studies have shown that a majority of word learning input in the natural environment is highly ambiguous, with only a small handful of super-informative instances. This characterization, though, may obscure details of what that “ambiguity” comprises, and how it nonetheless supports learning. We advance this idea by distinguishing a typology of three classes of word learning contexts: highly-informative (H) cases, ambiguous (A) cases which result in many possible hypotheses for the underlying meaning, and misleading (M) cases which generate largely just a single response, but one that is incorrect. We present two experiments on verb learning: first, we used a Human Simulation Paradigm (HSP) to categorize videos extracted from a parent-infant toy-play video corpus into these H/A/M classes. Second, we used the output of the HSP to select stimuli for a new cross-situational verb learning experiment—utilizing different combinations of H/A/M videos to assess their cumulative impact of learning.

Large language models (LLMs) can be reasonably thought of as models of idealized statistical learners. Thus, the extent to which they grasp the grammar of the language they are trained on suggests how much of it can be learned from memorization, abstraction, and generalization of linguistic input. However, the knowledge of LMMs’ grammar has largely been gleaned from examples of their outputs or datasets not designed to assess how native-like its knowledge is. In this study, we probed the knowledge of an LLM, GPT-3, with a graded grammatical acceptability task previously normed on humans. GPT-3’s ratings were correlated with human ratings, even with minimal examples. Moreover, GPT-3’s deviation from the human norms was predicted by the between-subject variation for each item, and these deviations were rarely outside of the range of human ratings. Follow-up analyses tested the extent to which local probabilistic structure drives these judgments using n-gram models.

Many paradigms in cognitive science posit that human learning is characterized by a limited capacity to represent the information relevant for a given task. We argue that excess capacity -- using more representational resources than needed for a task at hand -- is a plausible alternative paradigm for the study of human learning. Leveraging recent results from machine learning, we show that excess capacity can be consistent with high predictive ability. We also review extant empirical findings from the cognitive science literature, demonstrating that excess capacity learning can account for a range of empirical phenomena, such as humans' simultaneous yet apparently contradictory tendency to both memorize observations and capture higher-level patterns in them. We conclude by discussing promising directions for future inquiry under the excess capacity learning paradigm.

When people make decisions, these are often sequential, and later decisions frequently depend on the outcome of the previous decision. For instance, a doctor will first decide on a patient's treatment and then its duration. Grammar learning shows that humans can learn temporal regularities between such sequential information. However, little is known about sequential category learning tasks. Here, we investigate learning of regularities between two categorization tasks and generalization to novel objects. In an experiment, we varied whether a contingency between the outcome of two categorization tasks existed and whether they were adjacent (the tasks followed each other) or non-adjacent (an intervening task in between). Participants learned to categorize the objects of the second task in adjacent condition better than other conditions. However, both regularity conditions were beneficial to categorize novel objects (generalization). The results show the importance of considering temporal regularities between decision tasks in theories of category learning.

Environmental restrictions often permit the sampling of some items while excluding others. Such restrictions are termed sampling frames, whereby items can be selected based on their category membership (category frame), or possession of a target property (property frame). According to Bayesian principles, narrower property generalization is expected when a sample is subject to property sampling than category sampling. The current work examined whether sensitivity to such sampling frames could be increased through training with worked examples and practice. Experiment 1 found that training in property or category sampling enhanced sensitivity to that frame relative to a no-training control. Experiment 2 employed a pre-post design where all participants received training in both frames. A positive training effect was found, but only for those with a poor understanding of sampling frames on the pre-test. This work indicates the viability of appropriate training for increasing understanding of the implications of sample selection mechanisms.

Artificial grammar learning (AGL) experiments are commonly used to assess implicit learning. However, in these tasks participants may acquire explicit rather than implicit knowledge. We combined an AGL task with measures of metacognition to assess whether participants acquired explicit knowledge of the grammar. Participants were initially exposed to sequences of visual symbols generated by an artificial grammar, followed by a testing phase in which they classified sequences as either grammatical or ungrammatical. On a subset of trials (choice trials) participants were given the option to either take or skip this classification. If participants had explicit knowledge of the grammar, they should be more likely to take trials when they are confident and skip when they are not. Participants learned the grammar and reported explicit knowledge of the grammar. However, they performed similarly in forced and choice trials, suggesting that this metacognitive measure does not measure the explicitness of their knowledge of this grammar.

Ground Delay Programs (GDP) comprise the main interventions to optimize flight operations in congested air traffic networks. The core GDP objective is to minimize flight delays, but this may result in optimal outcomes for passengers — especially with connecting itineraries. This paper proposes an original passenger-centric approach to GDP by leveraging data on passenger itineraries in flight networks. First, we identify analytical drivers of passenger-centric operations in transportation systems. Second, we develop an integer program that balances flight delays and passenger delays in large-scale GDP operations. A rolling procedure decomposes the problem while ensuring global feasibility, enabling the model’s implementation in short computational times. Third, we propose statistical learning models to predict passenger itineraries and optimize GDP operations accordingly, enabling the model’s implementation when passenger itineraries are unknown by air traffic managers. Computational results based on real-world data suggest that our modeling and computational framework can reduce passenger delays significantly at small increases in flight delay costs, and that these benefits are robust to imperfect knowledge of passenger itineraries. Results highlight two major levers of passenger-centric operations: (i) delay allocation (which flights to delay vs. prioritize), and (ii) delay introduction (whether to deliberately hold flights to avoid passenger misconnections).

In many statistical learning problems, it is desired that the optimal solution conform to an a priori known sparsity structure represented by a directed acyclic graph. Inducing such structures by means of convex regularizers requires nonsmooth penalty functions that exploit group overlapping. Our study focuses on evaluating the proximal operator of the latent overlapping group lasso developed by Jacob et al. in 2009 . We implemented an alternating direction method of multiplier with a sharing scheme to solve large-scale instances of the underlying optimization problem efficiently. In the absence of strong convexity, global linear convergence of the algorithm is established using the error bound theory. More specifically, the paper contributes to establishing primal and dual error bounds when the nonsmooth component in the objective function does not have a polyhedral epigraph. We also investigate the effect of the graph structure on the speed of convergence of the algorithm. Detailed numerical simulation studies over different graph structures supporting the proposed algorithm and two applications in learning are provided. Summary of Contribution: The paper proposes a computationally efficient optimization algorithm to evaluate the proximal operator of a nonsmooth hierarchical sparsity-inducing regularizer and establishes its convergence properties. The computationally intensive subproblem of the proposed algorithm can be fully parallelized, which allows solving large-scale instances of the underlying problem. Comprehensive numerical simulation studies benchmarking the proposed algorithm against five other methods on the speed of convergence to optimality are provided. Furthermore, performance of the algorithm is demonstrated on two statistical learning applications related to topic modeling and breast cancer classification. The code along with the simulation studies and benchmarks are available on the corresponding author’s GitHub website for evaluation and future use.

Activity equations trying to mimic experimental catalytic performance derived from reaction profiles and microkinetic models have been the state of the art in modeling in the last decades. This approach has been able to reproduce semiquantitatively activity volcano plots leading to successful catalyst optimization through the use of descriptors. As systems become more complex (both catalysts and reactants), these methods face increasing limitations. Statistical Learning (SL) techniques can overcome these limitations and improve the search for descriptor-based performance equations. However, the black-box nature of SL techniques makes physical interpretation of the so-obtained models difficult. To advance in the integration of these methodologies to real problems, we have merged experimental activity and selectivity presented as a function of chemical descriptors from Density Functional Theory for the catalyzed hydrodehalogenation of CH2X2 (for X = Br, Cl) leading to three main products. The employed Bayesian procedure is able to identify robust equations for activity and selectivity as a function of only two descriptors. This work provides a starting point to solve complex reaction networks using a set of statistical learning tools and hybrid data., ACS Catalysis, 12 (2), ISSN:2155-5435

Oscillation-based testing of analogue electronic filters removes the need for test signal synthesis. Parametric faults in the presence of normal component tolerance variation are challenging to detect and diagnose. This study demonstrates the suitability of statistical learning and deep learning techniques for parametric fault diagnosis and detection by investigating several time-series classification techniques. Traditional harmonic analysis is used as a baseline for an in-depth comparison. Eight standard classification techniques are applied and compared. Deep learning approaches, which classify the time-series signals directly, are shown to benefit from the oscillator start-up region for feature extraction. Global average pooling in the convolutional neural networks (CNN) allows for Class Activation Maps (CAM). This enables interpreting the time-series signal’s discriminative regions and confirming the importance of the start-up oscillation signal. The deep learning approach outperforms the harmonic analysis approach on simulated data by an average of 11.77% in classification accuracy for the three parametric fault magnitudes considered in this work.